Can Australian law firms use AI to summarise matters safely?
Yes, with controls. NSW Supreme Court Practice Note SC Gen 23 expressly allows generative AI for summarising or reviewing documents and transcripts, but restricts what material can be entered. The 2024 statement from the Law Society of NSW, the Legal Practice Board of WA and the Victorian Legal Services Board + Commissioner says lawyers must protect confidentiality and personally verify AI output. That means a controlled platform, restricted-material rules, source-linked summaries and lawyer review.
Key takeaways
- SC Gen 23 permits Gen AI for summarising or reviewing documents and transcripts, subject to its restrictions on sensitive material (para 9A).
- Suppressed material, Harman-undertaking material and subpoenaed material must not go into Gen AI unless strict platform conditions are met.
- Plain transcription and translation tools are not "Gen AI" under the practice note (para 6); summarisation is.
- Regulators say lawyers cannot safely put confidential client information into public AI chatbots, and must personally verify AI-assisted work.
- Every case citation in AI-assisted submissions must be checked to exist, be accurate and be relevant, and not by AI alone.
Matter summaries suit AI: long transcripts, intake notes and bundles condensed into chronologies, issue lists and status updates. In NSW the rules are now specific. Here is what the court and the regulators say, and a workflow that protects confidentiality and accuracy.
What does NSW Practice Note SC Gen 23 allow?
The NSW Supreme Court's Practice Note SC Gen 23 (Use of Generative AI) was issued on 28 January 2025 and applies to proceedings from 3 February 2025. Three parts matter most for summaries:
- Para 6, the carve-out. Tools that only correct spelling or grammar, provide transcription or translation, or help with formatting, without generating substantive content, are not "Gen AI" for the practice note. A transcript is not covered; an AI-written summary of it is.
- Para 9B, permitted uses. Subject to para 9A, Gen AI may be used to generate chronologies, indexes and witness lists, and for summarising or reviewing documents and transcripts.
- Para 9A, the general prohibition. Certain material must not be entered into any Gen AI program unless strict conditions are met (see the table below).
The note also prohibits using Gen AI to generate the content of affidavits, witness statements and character references (para 10), so keep AI summaries away from evidence drafting beyond preparatory work.
What risks does the practice note identify?
Para 7 lists the limits practitioners should understand, including:
- "hallucinations": plausible but inaccurate or fictitious output, such as false citations and fabricated case or legislative references;
- dependence on underlying data that may be incomplete, out of date or not relevant to NSW or Australia;
- biased or inaccurate output;
- prompts being added to a model's database, unless disabled, and used to answer other users;
- a lack of safeguards for confidentiality, privacy or legal professional privilege when material is put into a public chatbot.
Para 8 adds that data entered into Gen AI programs may be used to train the model, potentially making confidential information available to others.
Which material must stay out of AI tools?
| Material | Rule under SC Gen 23 | Practical step |
|---|---|---|
| Information under non-publication or suppression orders | Not to be entered unless para 9A conditions are met | Flag at file level; exclude from AI by default |
| Material covered by the implied (Harman) undertaking | Same restriction | Tag discovery material on intake |
| Material produced on subpoena | Same restriction | Keep in a separate, excluded folder |
| Material subject to a statutory prohibition on publication | Same restriction | Partner sign-off before any AI use |
| Ordinary matter documents and your own transcripts | Summarising and review permitted, subject to 9A | Use a controlled platform; lawyer reviews output |
The para 9A conditions are demanding: the practitioner must be satisfied the information stays within the controlled environment of the platform, the supplier is bound by confidentiality restrictions so it is not made public or used to train any large language model, and it is used only in connection with that proceeding. In practice, many firms simply exclude these categories.
What do the legal regulators expect?
On 6 December 2024 the Law Society of NSW, the Legal Practice Board of Western Australia and the Victorian Legal Services Board + Commissioner issued a Statement on the use of artificial intelligence in Australian legal practice. It ties AI use to existing duties:
- Confidentiality. Lawyers cannot safely enter confidential, sensitive or privileged client information into public AI chatbots or other public tools, and should carefully review contract terms for commercial tools.
- Independent advice. AI tools cannot reason, understand or advise; the lawyer's own judgement is required.
- Competence and diligence. Summarisers cannot substitute for legal knowledge, and lawyers must personally verify AI-assisted documents.
- Fair costs. Bills should reflect the work actually done, and AI should not increase costs above traditional methods.
The statement also suggests risk-based policies on which tools are used, by whom and with what information, limiting AI to lower-risk tasks, and being transparent with clients. The Law Society of NSW's media release notes it applies to solicitors in NSW and to solicitors and barristers in Victoria and WA.
How should citations in AI summaries be checked?
For written submissions and summaries of argument prepared with Gen AI, SC Gen 23 (paras 16 and 17) requires the author to verify that all citations and legal references exist, are accurate and are relevant, and that verification must not be done solely with a Gen AI tool. Internal matter summaries are not submissions, but apply the same discipline: any authority an AI mentions is unverified until a person has opened the source.
A safe workflow for AI matter summaries
- Classify the material. Tag suppression, Harman, subpoena and statutory-prohibition material at intake and exclude it from AI processing.
- Use a controlled platform. Confirm in writing that inputs are not used for training, where data is processed and who can access it. Never use a public chatbot for client material.
- Transcribe, then summarise. Keep the verbatim transcript as the record; the summary is a working document.
- Require source references. Every point in the summary should cite a page, paragraph or timestamp so a reviewer can check it quickly.
- Lawyer review before filing. A practitioner checks the summary against the source and approves it before it goes to the matter file or the client.
- Check authorities outside the AI. Verify any case or legislation in a proper research database.
- Record and bill fairly. Note AI use on the file, and bill for the work actually done.
Where does a managed AI employee fit?
Sol is Wharq's role for law firms. In this kind of workflow it can transcribe and summarise calls and intake meetings, search the firm's documents and precedents before drafting, prepare a summary with source references, and add the follow-up tasks, with the draft going to a lawyer for review rather than straight to the file. It is connected only to the systems the role needs, and logs what it reads and writes. Wharq confirms hosting and data-handling specifics for your firm during the audit. See the roles on our services page, or read what a managed AI employee is.
Before connecting any client data, work through our AI security and supervision FAQ and the Privacy Act checklist for AI.
General information, not legal advice. Court practice notes differ between courts and jurisdictions and are updated; check the current rules for each court and your professional obligations before relying on this summary.
AI case summaries: frequently asked questions
Is AI transcription covered by SC Gen 23?
Not by itself. Paragraph 6 says technology that only provides transcription or translation, corrects spelling or grammar, or helps with formatting, without generating substantive content, is not Gen AI for the practice note. Summarising a transcript with a generative tool is covered.
Can we put subpoenaed documents into an AI tool?
Only if the paragraph 9A conditions are met: the material stays in a controlled environment, the supplier is bound by confidentiality and does not use it to train models, and it is used only for that proceeding. Many firms exclude this material entirely.
Can we use ChatGPT for client matter summaries?
Not a public version with confidential client information. The 2024 regulators' statement says lawyers cannot safely enter confidential, sensitive or privileged information into public AI chatbots. Commercial tools need their contract terms reviewed first.
Do we have to tell clients we use AI?
The 2024 statement encourages transparency, including disclosing to clients when and how AI was used in a matter and how it is reflected in costs if the client asks. Court rules may also require disclosure in some documents.
Does SC Gen 23 apply outside NSW?
It applies to proceedings in the NSW Supreme Court. Other courts have their own practice notes or guidance, so check the rules of each court you appear in.
Sources
- Supreme Court of NSW, "Practice Note SC Gen 23: Use of Generative Artificial Intelligence" (issued 28 January 2025, commenced 3 February 2025)
- Victorian Legal Services Board + Commissioner, "Statement on the use of artificial intelligence in Australian legal practice" (published 6 December 2024)
- Law Society of NSW, "AI guidance to safeguard consumers of legal services" (accessed October 2026)